Parallelizing Raytracing

Gillian SmithCMPE 220

February 19th, 2008

Agenda What is raytracing? How raytracers work Parallelization

What Is Raytracing? 3D rendering

technique Mathematical

representation of scenes

Extremely computationally intensive

Results capture complex lighting effects

What Is Raytracing?

What Is Raytracing?

How Raytracers Work Image plane is

placed between the camera and the scene

The color of each pixel in that plane is computed by the raytracer

Pixels are independent of each other

Algorithm Overview (Per Pixel)1. Shoot a ray from the camera through the

pixel and towards the scene2. Compute the intersection between

objects in the scene and the ray3. For the closest intersection point to the

camera, compute the color that should be stored in the pixel

Compute Ray through Pixel

Front ViewSide View

P0: Camera’s Position

D: Camera’s Direction

U: Camera’s “Up” Direction

U

D

Pi

Θ

Ray: R = P0 + tV

|| 0

0

PP

PPV

i

i

Algorithm Overview (Per Pixel)1. Shoot a ray from the camera through the

pixel and towards the scene2. Compute the intersection between

objects in the scene and the ray3. For the closest intersection point to the

camera, compute the color that should be stored in the pixel

Compute Intersection

Slide Credit: Greg Humphreys, University of Virginia

P0

V

Sphere: |P - C|2 – r2 = 0

Point on Ray: P = P0 + tV

C

P

P’

R

P0

V

Compute IntersectionSubstitute for P in equation for sphere:

|P0 + tV – C|2 – r2 = 0

Solve for t, choose smaller value

C

P

P’

R

P0

V

Algorithm Overview (Per Pixel)1. Shoot a ray from the camera through

the pixel and towards the scene2. Compute the intersection between

objects in the scene and the ray3. For the closest intersection point to

the camera, compute the color that should be stored in the pixel

Determine color at pixel Raycasting method

Look up color from scene specification

Raytracing method Find contribution to

pixel from each light source

Reflections Refractions

Local Illumination Light has a color, LC Object’s material

Diffuse color, CD Specular color, CS Specular highlight, S

Pixel Diffuse CDLC(N.L)

Pixel Specular CSLC(R.V)S

Final color is sum of all diffuse and specular colors from every light in the scene

Global Illumination Illumination effects dependent on other

objects in the scene Shadows Reflections Refraction

Calculated by sending additional rays from the object Shadows: send ray to light Reflections: reflect ray around normal Refraction: depends on index of refraction

Computational Expense•Rendered at 800x600

•PowerBook G4 (1.25GHz PowerPC, 1GB RAM)

•> 10 hours to render

Computational Expense•598x634 image.

•131,072 triangles

•Depth of field effect

•GeForce 7800 GTX

•Less than 2 minutes.

Raytracing Can Be Parallelized “Embarrassingly (pleasantly?) parallel” Pixels are independent of each other All pixels must known complete scene

specification SPPM

Computing global illumination involves recursion

Some pixel computations will finish before others

Project: Kestrel Implementation Additional math

tangent square root

Precision 32bit fixed point Scene must stay in a particular range

Space Memory: Store scene in every PE Register: Lots of values to keep track of for

each computation

Parallelizing Raytracing

Gillian SmithCMPE 220

February 19th, 2008